Accurate large-scale simulations of siliceous zeolites by neural network potentials
Andreas Erlebach, Petr Nachtigall, Lukáš Grajciar
Abstract
Abstract The computational discovery and design of zeolites is a crucial part of the chemical industry. Finding highly accurate while computational feasible protocol for identification of hypothetical siliceous frameworks that could be targeted experimentally is a great challenge. To tackle this challenge, we trained neural network potentials (NNP) with the SchNet architecture on a structurally diverse database of density functional theory (DFT) data. This database was iteratively extended by active learning to cover not only low-energy equilibrium configurations but also high-energy transition states. We demonstrate that the resulting reactive NNPs retain DFT accuracy for thermodynamic stabilities, vibrational properties, as well as reactive and non-reactive phase transformations. As a showcase, we screened an existing zeolite database and revealed >20k additional hypothetical frameworks in the thermodynamically accessible range of zeolite synthesis. Hence, our NNPs are expected to be essential for future high-throughput studies on the structure and reactivity of siliceous zeolites.